In college I had a professor assign us to write a 100% plagiarized paper. You had to highlight every word in a color associated with the source. You couldn't plagiarize more then one sequential sentence from a single source.
It ended up being harder then writing an ordinary paper but taught us all a ton about citation and originality. It was a really cool exercise.
I imagine something similar could be done to teach students to use AI as a research tool rather then as a plagiarization machine.
Reminds me of that thing Olmo3 has in the web demo, where it shows you a trace for a selected sentence generated by the LLM and points you at the exact document in the training data it comes from verbatim. If they all had that it would be really trivial to verify sources, and more or less does that exact exercise for you at a press of a button.
Contemporary alternative: Copy/paste an essay entirely from LLM output, but make sure none of the information contained checks out. One would want to use an older model for this. :-)
It ended up being harder then writing an ordinary paper but taught us all a ton about citation and originality. It was a really cool exercise.
I imagine something similar could be done to teach students to use AI as a research tool rather then as a plagiarization machine.